This study tested the hypothesis that some MRI-based radiomic features could lose their texture descriptive power below a volume threshold, compromising the robustness of prediction models. The ability of features to discriminate different textures as a function of the tumor volume was investigated on T2-weighted images of a customized pelvic phantom. The images were acquired on three scanners and considered three different textures, from a finer to a coarser one. The texture discriminative ability was shown to depend on tumor volume, with most features losing this property with a VOI smaller than 1 cm3 at 1.5 T.
This abstract and the presentation materials are available to members only; a login is required.